harness-code-review

Automated code review uses mechanical checks and graph-scoped context to evaluate pull requests. It coordinates parallel subagents for compliance, bug detection, security, and architecture to produce a unified, deduplicated findings report.

Graph-scoped context in code review bundles context per domain to keep reviews focused. This mechanism ensures that automated findings are evaluated against the relevant architectural constraints and dependencies, reducing manual effort and errors.

Deduplicate findings from security scans and bug detection by running parallel subagents that cross-reference mechanical checks. The pipeline filters overlapping results against evidence-based findings to produce a unified report.

Automated code review outputs structured findings designed for CI handoffs and inline PR comments. The evidence-rich output includes justification citations and remediation suggestions formatted for direct integration into review workflows.

Automated code review integrates mechanical checks like linting, type checks, and tests alongside security scans. It also supports optional threat modeling for deeper analysis, coordinating these phases to enforce architectural constraints.

Automated multi-phase code review relies on graph-scoped context bundles, meaning review focus depends on domain scoping accuracy. While it deduplicates mechanical scans, optional threat modeling requires deeper analysis beyond standard automated validation.